Research On Fusion Approach Of Conflicting Uncertainty Information And Its Application In Fault Diagnosis

In practical application of Dempster-Shafer evidence theory, the combination method of the conflicting uncertainty information is not yet perfect, which mainly reflects on three aspects:the first is how to describe the uncertainty of information and estimate the reliability of information sources more effectively; the scond is how to express the degree of conflict in case of conflict between multi-source information and evaluate the applicability of classic Dempster combination rule of evidence more effectively; the third is how to resolve the counter-intuitive result of the classic Dempster combination more effectively and improve the effectiveness of uncertain information fusion. These issues have important practical applications and theoretical research value in equipment fault diagnosis, image processing, traffic intelligent control, etc.Main work and achievements of the thesis are as follows:(1) A new method of description of uncertainty information and combination of evidence based on total uncertainty is proposed.Since the descriptions of uncertainty of information by the several current uncertainty measurement functions in evidence theory are incomplete, a concept of the combined total uncertainty which can be obtained by combining the uncertainty degrees of two BOEs on Cartesian product is proposed. It is proved that the total uncertainty of a body of evidence (BOE) can satisfy the five conditions for being a description of uncertainty of information, which are probability consistency, set consistency, measure range, subadditivity, and additivity. The degrees of the total uncertainty of BOEs can be used to estimate the reliability of the information sources. According to both the principle of maximum entropy and the weighted-average method of combination, an improved combination method based on the total uncertainty degree of every BOE is proposed. In the proposed method, firstly, the weighting factors for modifying the original BOEs are obtained by using ratio values of the total uncertainty of each BOE and the combined total uncertainty on Cartesian product, and then the BOEs modified by the weighting factors are combined by using the classical Dempster’s rule of combination. Results of numerical examples of information fusion show that the proposed method has better applicability and effectiveness compared with the existing methods.(2) A new distance function between BOEs based on supporting probability distance function is proposed, and then combining with the new distance and classical conflict coefficient k, a novel criterion to evaluate the applicability of Dempster’s rule of combination is presented.In order to solve the limitations of some common distances, such as Jousselme’s distance and the pignistic probability distance between BOEs, a new probability function which is called the supporting probability function, is proposed with explanation of its mathematical meaning and properties. Comparative analysis with Pignistic probability function was carried out to demonstrate the better characteristics of the new function. Also, a new distance function, which is called the supporting probability distance, is defined. Combining with the supporting probability distance and the classical conflict coefficient k, a new criterion for evaluating the applicability of Dempster’s rule is proposed. The proposed criterion evaluates whether Dempster’s rule can be used to combine a given group of BOEs which may contain conflicting information. Analytical results of numerical examples show that the proposed evaluation method has better versatility and robustness compared with the existing methods.(3) A new method of combination based on the supporting probability distance is proposed.Aim at the combination of conflicting BOEs, a new combination method based on the supporting probability distance is proposed. Firstly, the credibility and discounting factor of each BOE are generated by the supporting probability distance. Secondly, the weighted BOEs are combined using Dempster’s rules. Some numerical examples were given to show the effectiveness of the proposed method.(4) A new fault diagnosis method with decision-level fusion based on the supporting probability distance is proposed.A new fault diagnosis method with decision-level fusion based on the supporting probability distance is proposed by a synthesis of the research results in this paper, such as the method for measuring the conflict between BOEs, the new criterion for evaluating the applicability of classical Dempster rule of combination, and the new method of combination based on the supporting probability distance. The main procedures of the new fault diagnosis method are: firstly, collect information from different sensors and build BOEs for each fault feature; secondly, evaluat the conflict between BOEs and make a combination of the BOEs; finally, make a decision of diagnosis according to the combination results. The diagnosis results of the experiment on a flexible rotor showed that the proposed method can enhance the diagnostic accuracy and reliability.